I am training a cnn segmentation model and I need some analog of F1 score enter image description here

So, we have GT as red rectangles (called "red") and Pred as blue rectangles (called "blue"). It is clear that there is no FP (all blue have 50+% coverage by red) and there is one FN (right red is covered by 50-% of blue). I can compute accuracy and recall: 100% and 50%. But I cannot compute F1 score because I have different values of TP for blue and red. What is the best way to adopt F1 score for this case?

Normally this is done by IoU, but in my case I need an analog of F1 score.



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.